Development of a Clinical Framework-Anastomotic Leakage Prediction Score (CF-ALPS) After Colorectal Surgery

Background: Anastomotic leakage (AL) is a severe complication of colorectal surgery and is associated with high morbidity, mortality, and prolonged hospital stay. Current predictive models vary in complexity and utility, highlighting the need for clinically accessible and accurate tools. This study aimed to develop and validate the clinical framework-anastomotic leakage prediction score (CF-ALPS) score, a practical and accessible risk prediction model for AL that integrates patient-, tumor-, and surgery-related factors. Methods: A retrospective cohort of 294 patients who underwent colorectal surgery between 2019 and 2024 was analyzed. Patients were categorized into the AL (n = 84) and non-AL groups (n = 210). The factors included age, sex, hypoalbuminemia, and comorbidities. Tumor-related factors included lymph node stage and neoadjuvant therapy, while surgical variables included urgency, duration, and type of procedure. The outcomes evaluated were the incidence of AL, duration of hospital stay, and in-hospital mortality. Independent predictors were identified using multivariate logistic regression analysis. The CF-ALPS score, which was developed from significant predictors, was validated using ROC curve analysis and 10-fold cross-validation. Results: A total of 294 patients who underwent colorectal surgery were included, of whom 84 (28.57%) developed AL. A male predominance was observed in the AL group (73.81% vs. 36.19%; p = 0.001). Nutritional status played a critical role, with significantly lower albumin levels in AL patients (2.8 +/- 0.5 g/dL vs. 3.5 +/- 0.4 g/dL; p < 0.001). Independent predictors of AL included hypoalbuminemia (<3.0 g/dL, OR: 0.52, p < 0.001), ASA score (OR, 1.85; p = 0.004), advanced lymph node stage (N2/N3, OR: 1.94, p = 0.037), neoadjuvant therapy (OR, 2.89; p = 0.002), and emergent surgery (OR, 1.67; p = 0.042). These variables formed the basis of the CF-ALPS score, which assigns weighted points based on the magnitude of their ORs. The CF-ALPS model achieved a ROC AUC of 0.82 (95% CI: 0.75-0.89) with a sensitivity of 85.0% and specificity of 78.0%. A cutoff score >= 7 demonstrated optimal risk stratification, classifying patients into high- and low-risk groups with a positive predictive value (PPV) of 72.0% and a negative predictive value (NPV) of 88.0%. Cross-validation yielded a moderate AUC of 0.44 (SD = 0.062). Conclusions: The CF-ALPS score offers a simple and effective tool for AL risk prediction in colorectal surgery, emphasizing its practicality and clinical integration. Although its predictive accuracy was moderate, further prospective multicenter validation is warranted.